In econometrics, numbers only begin to speak when you anchor them to something lived—fuel bought at dusk, wages earned under a humid sky, prices that rise a little too quietly. Let’s take a few small, grounded datasets—simple, imperfect, but real enough to carry meaning—and walk the equations into daylight.
Example 1: Education and Income (Cross-Sectional Data)
Imagine a small survey from households around Mombasa:
| Years of Schooling (x) | Monthly Income (KES ‘000) (y) |
|---|---|
| 8 | 18 |
| 10 | 22 |
| 12 | 30 |
| 14 | 36 |
| 16 | 45 |
After estimation, suppose we get:
ŷ = -5 + 3x
How to read this, carefully:
- Each extra year of schooling adds about KES 3,000 to monthly income.
- The negative intercept is nonsense in real life—no one earns negative income. It’s a reminder: models extrapolate beyond dignity.
Quiet doubt: Is schooling causing income—or standing in for family background, networks, or luck?
Example 2: Inflation and Food Prices (Time Series)
Take monthly maize flour prices across Kenya:
| Month | Price (KES) |
|---|---|
| Jan | 120 |
| Feb | 125 |
| Mar | 130 |
| Apr | 138 |
| May | 150 |
Suppose:
ŷt = 10 + 0.9yt-1
Interpretation:
- Prices today depend heavily on yesterday (β ≈ 0.9).
- Shocks fade slowly—once prices rise, they tend to stay risen.
But pause: Where are droughts? Transport costs? Policy shocks? The equation is calm; reality is not.
Example 3: Omitted Variable Bias (The Hidden Distortion)
Return to income and education—but now add experience (z).
True model:
y = β₀ + β₁x + β₂z + ε
If you ignore experience:
β̃₁ = β₁ + β₂ · Cov(x,z) / Var(x)
What this means in plain terms:
- If educated people also tend to be more experienced, your model overstates the return to education.
- You think schooling pays more than it truly does.
A small omission, a large distortion. This is where many confident conclusions quietly collapse.
Example 4: Testing Significance (Is It Real or Noise?)
From Example 1, suppose:
- Estimated slope: 3
- Standard error: 0.8
t = 3 / 0.8 = 3.75
Interpretation:
- This is statistically significant.
- But significance is not importance. A precise estimate can still describe a trivial or misunderstood relationship.
Example 5: Instrumental Variables (A Fragile Rescue)
Suppose schooling is endogenous. You use distance to school (z) as an instrument.
- Cov(z, y) = -6
- Cov(z, x) = -2
β̂IV = 3
Same estimate—but earned differently.
The uncomfortable question: Does distance affect income only through education? Or does it also reflect rural disadvantage, infrastructure gaps, forgotten regions?
If the instrument is flawed, the elegance of the equation becomes a disguise.
Closing Reflection
These examples are small—almost humble. But that’s the point. Econometrics was never meant to dominate reality, only to negotiate with it.
In places like Mombasa, where economies shift with tides, tourism, and trade winds, the data will always be thinner than the truth it tries to hold.
So treat each equation as a lens, not a verdict. It sharpens your view—but it never shows the whole landscape.
No comments:
Post a Comment